Prognostic Value of miR-21 in Various Cancers: An Updating Meta-Analysis

Background Recently, more and more studies investigated the value of microRNA (miRNA) as a diagnostic or prognostic biomarker in various cancers. MiR-21 was found dysregulated in almost all types of cancers. While the prognostic role of miR-21 in many cancers has been studied, the results were not consistent. Methods We performed a meta-analysis to investigate the correlation between miR-21 and survival of general cancers by calculating pooled hazard ratios (HR) and 95% confidence intervals (CI). Results The pooled results of 63 published studies showed that elevated miR-21 was a predictor for poor survival of general carcinomas, with pooled HR of 1.91 (95%CI: 1.66–2.19) for OS, 1.42 (95% CI: 1.16–1.74) for DFS and 2.2 (95% CI: 1.64–2.96) for RFS/CSS. MiR-21 was also a prognostic biomarker in the patients who received adjuvant therapy, with pooled HR of 2.4 (95%CI: 1.18–4.9) for OS. Conclusions Our results showed that miR-21 could act as a significant biomarker in the prognosis of various cancers. Further studies are warranted before the application of the useful biomarker in the clinical.


Introduction
Due to the aging and growth of population as well as an increasing adoption of cancer-related lifestyle such as smoking and ''westernized'' diets, cancer has been a major public health problem all around the world [1]. Almost one in four deaths in the United States is related with cancer in 2012 [2]. Lack of efficiently diagnostic and prognostic biomarkers is responsible for the high mortality rates caused by cancer [3].
MicroRNAs (miRNAs), approximately 22 nucleotides in length, are a class of highly conserved RNAs that negatively regulate gene expression at post-transcriptional level by base pairing with the 39untranslated region of target mRNAs, resulting in either mRNA degradation or translational inhibition [4,5]. Many studies have demonstrated that miRNAs play important roles in various biological processes, such as cellular development, differentiation, proliferation, cell death, angiogenesis and metabolism [6][7][8][9]. The success of utilizing miRNAs as diagnostic or prognostic markers from expression profiling has been reported in many studies.
MiR-21 was one of the most frequently studied cancer-related miRNAs and dysregulated in most cancers by acting as oncogene [10][11][12][13][14]. Up-regulated miR-21 could increase tumor growth, metastasis and invasion and reduce sensitivity to chemotherapy by its various targets [15][16][17][18]. Cancer patients with higher expression of miR-21 always had a worse prognostic outcome. But some studies represented inconsistent or even opposite results, such as the study of Valladares-Ayerbes et al. [19]. So we performed this meta-analysis to reveal the prognostic value of miR-21 in various cancers.
We also carefully examined the references of articles and reviews to explore potentially additional studies. Studies were eligible if they met the following criteria: (a) studied patients with any type of cancers; (b) expression of miR-21 was measured; (c) the association between expression of miR-21 and clinical outcome was investigated; (d) full text articles in English. Studies were excluded based on the following criteria: (a) reviews, letters or laboratory studies; (b) studies had overlapping or duplicate data; (c) absence of key information for further analysis [20].

Data extraction
Data were evaluated and extracted independently from the eligible studies by two investigators (Zhou and Wang) under the guidelines of a critical review checklist of the Dutch Cochrane Centre proposed by Meta-analysis of Observational Studies in Epidemiology (MOOSE) [21]. The following items were recorded: first author's name, year of publication, country or area of origin, ethnicity, cancer type, sample type, TNM stage, method, total number of patients, cut-off value, follow ups and HRs of miR-21 for overall survival (OS), disease-free survival (DFS), recurrencefree survival (RFS) or cancer specific survival (CSS) with their 95% confidence intervals (CIs) and P value. If not available, data were extracted by the method of Tierney et al. [20]. When discrepancies existed between the two investigators, another investigator (Huang) was invited to discuss until a consensus was reached.

Statistical analysis
All the HRs with their 95% CIs were used to calculate pooled HRs. Cochran's Q test and Higgins I-squared statistic were used to check the heterogeneity of pooled results. A P,0.10 for Q-test suggested significant heterogeneity among studies, and the random-effects model (DerSimonian-Laird method) was applied to calculate the pooled HRs [22]. Otherwise, the fixed-effects model (Mantel-Haenszel method) was used [23]. Begg's funnel plot and the Egger's linear regression test were conducted to evaluate publication bias of literatures and a p,0.05 was considered significant [24]. Trim and fill method was applied to assess potential asymmetry in the funnel plot. Statistical analyses were performed in STATA software version 12.0 (STATA Corporation, College Station, TX, USA). All P values were two-sided.

Study characteristics
After careful read and selection, a total of 63 articles [19, were retrieved according to the inclusion and exclusion criteria. 55 of 63 articles investigated the prognostic role of miR-21 for OS, 17 for DFS, 8 for RFS and 3 for CSS. Schetter et al. [25], Hwang et al. [40] and Akagi et al. [79] presented separate HR by different ethnic background; Mathe et al. [33]_ENREF_34, Liu et al. [51], Toiyama et al. [80], Nielsen et al. [44] and Markou et al. [82] investigated the role of miR-21 in different type of samples; Voortman et al. [43] reported results from two centers. So the data from these studies were considered separately in our analysis. As there were only 3 studies for CSS, we combined the results for CSS and RFS together as RFS/CSS. Thus, a total of 63 studies including 6720 patients evaluating OS, 19 studies including 1965 cases for DFS and 11 studies including 1696 patients for RFS/ CSS were considered in this analysis. The detailed screening process was shown in Figure 1.
The main characteristics of eligible studies were listed in Table 1. Ethnicity background of patients were classified as Asian, Caucasian and mixed populations. Cancer types of cases were various, among which lung cancer, pancreatic cancer and gastrointestinal (GI) cancers were mostly investigated. Tissue samples including Frozen or formalin-fixed and paraffin-embedded (FFPE) tissues were used in 53 studies, while 11 studies used circulation samples (plasma, serum or blood) and one study by Ota et al. [47] applied bone marrow samples. Quantitative real-time PCR (qRT-PCR) was widely used in 57 studies and in situ hybridisation (ISH) assay was used in the other 6 studies. The most frequently used cut-off value was the median which was applied in 26 studies and the other values ranged from the mean to the highest quarter value.

Outcomes from eligible studies
The main results of this meta-analysis are shown in Table 2. For 63 studies evaluating OS for miR-21, we found high expression of miR-21 predicting a worse outcome with the combined HR of 1.91 (95%CI: 1.66-2.19; P heterogeneity ,0.001; Figure 2). Similarly predictive roles of miR-21 for DFS and RFS/CSS were also investigated with pooled HR of 1.42 (95% CI: 1. 16 A total of seven studies [27,39,40,43,53,82] investigated the prognostic role of miR-21 in the patients who received adjuvant therapy which yielded a significantly pooled HR of 2.4 (95%CI: 1.18-4.9; P heterogeneity ,0.001).

Publication bias
Begg's funnel plot and the Egger's linear regression test were used to assess publication bias. However, the funnel plots were asymmetric and the P values of Egger's test for OS, DFS and RFS/CSS were ,0.001, 0.011 and 0.003, respectively. Thus, a trim and fill method was conducted and pooled HRs were recalculated with hypothetically non-published studies to evaluate the asymmetry in the funnel plots. The recalculated HRs did not change significantly for OS (HR = 1.61, 95%CI: 1.41-1.83; P heterogeneity ,0.001; Figure 3) and RFS/CSS (HR = 2.01, 95%CI: 1.54-2.77; P heterogeneity = 0.018). But the prognostic role of high expression of miR-21 for DFS was weaken with a recalculated HR of 1.11 (95%CI: 0.9-1.38; P heterogeneity ,0.001).

Discussion
MiR-21, a well-known onco-miR, is up-regulated in most malignancies. Acting on various target genes such as PTEN [87] and PDCD4 [18], miR-21 plays an important role in the process of cell proliferation, migration, invasion, drug resistance [88] and so on. It has been reported that miR-21 could regulate Ras/ MEK/ERK pathway so to influence the tumor formation. Moreover the incidence of lung tumors is higher in miR-21 overexpression mice, while lower in miR-21 knockout mice [89]. Additionally, miR-21 has been proposed as a marker of cancers for diagnosis in circulation [90,91], stool [92] and sputum [93], prediction in therapy response [59] and prognosis of patients.
Nair et al. [94] systematically reviewed and synthesized that miRNAs showed promising associations with outcomes of various cancers. As the first meta-analysis [95] of miR-21 related to outcomes of various cancers, Fu et al. retrieved 17 studies and found higher level of miR-21 might be associated with poorer clinical outcome, especially in subgroup of head and neck squamous cell carcinoma and digestive carcinoma. Recently, Wang et al. [96] analyzed the value of circulating miR-21 and yielded a conclusion that circulating miR-21 might act as a significantly prognostic biomarker but not be suitable for a sensitive diagnostic biomarker. However, the number of studies included in these analyses was relatively small and the obtained results might not be powerful. In terms of this, we performed this updated meta-analysis including 63 articles and demonstrated that high expression of miR-21 was a significant marker for predicting worse outcomes of various cancers (HR was 1.91, 2.2 and 1.42 for OS, RFS/CSS and DFS, respectively). Subgroup analyses revealed that high expression of miR-21 could predict a worse OS in GI tumors, pancreatic cancer, lung cancer, breast cancer and liver cancer, a worse DFS in pancreatic cancer and lung cancer and poor RFS/CSS in GI tumors, lung cancer and prostate cancer. Regardless of the ethnicity background or sample source, high expression level of miR-21 was a significantly negative prognostic marker for various malignancies. As publication bias was observed, a trim and fill method was adopted to calculate the adjusted HRs. The results for OS and RFS/CSS did not change, but the results for DFS were altered.
Recently, many studies demonstrated that miRNAs including miR-21 had great potential as biomarkers for various cancers. However, several problems should be well solved before utilizing them as diagnostic or prognostic biomarkers in the clinical. As is known, non-invasive circulation sample (plasma/serum) or body fluid sample could be obtained more conveniently than tissue sample. However, studies using different types of samples may yield different results [51]. Tsujiura et al. [91] found that some individuals might even have opposite tendency of the expression levels of miRNAs in tumor tissue and plasma. Now, many studies have investigated the clinical impact of miRNAs from exosomes which were small membrane vesicles containing proteins and nucleotides [97]. In our study, it is pleasing that high expression of miR-21 in the tissue (FFPE/frozen tissue) or circulation both predicted poor outcomes. Thus, we might assume that patients with high expression of miR-21 from any type of sample might suffer worse clinical outcomes. Yet, normalization among different studies was not consistent. The internal controls used for tissue samples are relatively consistent ranging from U6 to U44, while there is no consensus on suitable small RNA reference genes for circulation or body fluid sample. MiR-16 was used as a reference gene in some studies [66,78]. But the optimal way for miRNA normalization in circulation or body fluid sample is probably the spiked-in normalization method [98]. Therefore, future studies focusing on the consistent normalization are warranted. In addition, as biomarkers, a panel of miRNAs might be more sensitive and specific than a single miRNA [99,100]. The combination of miR-21 and some specific miRNAs might elevate its predictive power. Finally, methods for detecting miRNAs were diverse, among which RT-PCR was one of the most widely used approaches. Nevertheless many new methodologies emerged, such as the next-generation sequencing approach [101] and the electrochemical approach [102]. In short, a proper method for clinical application should be less expensive, reproducible, stable and with high sensitivity and specificity. Accordingly, great efforts should be made in the future to apply miRNAs including miR-21 as reliable biomarkers in the clinical.
Several limitations of this study should be considered. First, the studies retrieved in our study were limited in English, which might partially contribute to the observed publication bias. By conducting the trim and fill method, we found that the pooled results did not change significantly except for DFS. Thus, attention should be paid to the prognostic role of miR-21 for DFS. Second, different countries, cancer types, methods and other variables might contribute to the relatively large heterogeneity in this study. Third, the number of studies investigating some special types of cancer was small. For instance, there was only one study focusing on mesothelioma [30]. More studies on these cancers are needed in the future.
In conclusion, the evidence from the meta-analysis revealed that high expression level of miR-21 was a negative predictor for survival in various cancers, especially for OS and RFS/CSS. However, our results should be considered with caution due to the limitations listed above. To better understand and use miRNAs as biomarkers in the clinical, more large-scale and standard investigations are worth conducting.

Supporting Information
Checklist S1 PRISMA Checklist.